Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Interpretable Vision Transformer as a Fingerprint-Based Diagnostic Aid for Kabuki and Wiedemann-Steiner Syndromes

Marilyn Lionts, Arnhildur Tomasdottir, Viktor I. Agustsson, Yuankai Huo, Hans T. Bjornsson, Lotta M. Ellingsen

TL;DR

This study addresses the diagnostic gap for Kabuki syndrome (KS) and Wiedemann-Steiner syndrome (WSS) by leveraging dermatoglyphic signals through a vision transformer (ViT). A fingerprint dataset from KS, WSS, and controls was collected, preprocessed with quality filtering and Gabor-based enhancement, and analyzed with an ensemble ViT classifier augmented by attention heatmaps to distinguish KS, WSS, and controls, achieving AUCs up to 0.85. The results indicate syndrome-specific fingerprint features beyond persistent fetal pads and demonstrate a noninvasive, interpretable diagnostic aid that could expand access to genetic diagnostics, including potential smartphone-based capture for wider use. This work lays groundwork for dermatoglyphic AI tools in rare genetic diseases and highlights interpretability as a key component for clinical adoption.

Abstract

Kabuki syndrome (KS) and Wiedemann-Steiner syndrome (WSS) are rare but distinct developmental disorders that share overlapping clinical features, including neurodevelopmental delay, growth restriction, and persistent fetal fingertip pads. While genetic testing remains the diagnostic gold standard, many individuals with KS or WSS remain undiagnosed due to barriers in access to both genetic testing and expertise. Dermatoglyphic anomalies, despite being established hallmarks of several genetic syndromes, remain an underutilized diagnostic signal in the era of molecular testing. This study presents a vision transformer-based deep learning model that leverages fingerprint images to distinguish individuals with KS and WSS from unaffected controls and from one another. We evaluate model performance across three binary classification tasks. Across the three classification tasks, the model achieved AUC scores of 0.80 (control vs. KS), 0.73 (control vs. WSS), and 0.85 (KS vs. WSS), with corresponding F1 scores of 0.71, 0.72, and 0.83, respectively. Beyond classification, we apply attention-based visualizations to identify fingerprint regions most salient to model predictions, enhancing interpretability. Together, these findings suggest the presence of syndrome-specific fingerprint features, demonstrating the feasibility of a fingerprint-based artificial intelligence (AI) tool as a noninvasive, interpretable, and accessible future diagnostic aid for the early diagnosis of underdiagnosed genetic syndromes.

An Interpretable Vision Transformer as a Fingerprint-Based Diagnostic Aid for Kabuki and Wiedemann-Steiner Syndromes

TL;DR

This study addresses the diagnostic gap for Kabuki syndrome (KS) and Wiedemann-Steiner syndrome (WSS) by leveraging dermatoglyphic signals through a vision transformer (ViT). A fingerprint dataset from KS, WSS, and controls was collected, preprocessed with quality filtering and Gabor-based enhancement, and analyzed with an ensemble ViT classifier augmented by attention heatmaps to distinguish KS, WSS, and controls, achieving AUCs up to 0.85. The results indicate syndrome-specific fingerprint features beyond persistent fetal pads and demonstrate a noninvasive, interpretable diagnostic aid that could expand access to genetic diagnostics, including potential smartphone-based capture for wider use. This work lays groundwork for dermatoglyphic AI tools in rare genetic diseases and highlights interpretability as a key component for clinical adoption.

Abstract

Kabuki syndrome (KS) and Wiedemann-Steiner syndrome (WSS) are rare but distinct developmental disorders that share overlapping clinical features, including neurodevelopmental delay, growth restriction, and persistent fetal fingertip pads. While genetic testing remains the diagnostic gold standard, many individuals with KS or WSS remain undiagnosed due to barriers in access to both genetic testing and expertise. Dermatoglyphic anomalies, despite being established hallmarks of several genetic syndromes, remain an underutilized diagnostic signal in the era of molecular testing. This study presents a vision transformer-based deep learning model that leverages fingerprint images to distinguish individuals with KS and WSS from unaffected controls and from one another. We evaluate model performance across three binary classification tasks. Across the three classification tasks, the model achieved AUC scores of 0.80 (control vs. KS), 0.73 (control vs. WSS), and 0.85 (KS vs. WSS), with corresponding F1 scores of 0.71, 0.72, and 0.83, respectively. Beyond classification, we apply attention-based visualizations to identify fingerprint regions most salient to model predictions, enhancing interpretability. Together, these findings suggest the presence of syndrome-specific fingerprint features, demonstrating the feasibility of a fingerprint-based artificial intelligence (AI) tool as a noninvasive, interpretable, and accessible future diagnostic aid for the early diagnosis of underdiagnosed genetic syndromes.
Paper Structure (8 sections, 5 figures, 1 table)

This paper contains 8 sections, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Method
  3. Data Collection
  4. Model Architecture
  5. Experiments
  6. Results
  7. Conclusion and Discussion
  8. Acknowledgements

Figures (5)

  • Figure 1: Three representative fingerprint images acquired using the GenePrint mobile application and an external optical scanner. The images correspond to a single finger from patients with KS (left), WSS (center), and a control individual (right)..
  • Figure 2: Fingerprint image acquired using the GenePrint mobile application and external optical scanner (left) and the same image after preprocessing with the Gabor filter–based enhancement algorithm (right).
  • Figure 3: Overview of the fingerprint-based classification pipeline. Fingerprint data were collected from participants with Kabuki syndrome (KS), Wiedemann-Steiner syndrome (WSS), and unaffected controls using a standardized mobile app and an optical fingerprint scanner. Fingerprint images were subsequently preprocessed and filtered according to image quality. A Vision Transformer model was trained and evaluated across three classification tasks using ensemble predictions. Attention heatmaps were generated to visualize model focus and enhance interpretability.
  • Figure 4: Receiver Operating Characteristic (ROC) curves for the three classification tasks, generated from ensemble predictions on the held-out test set: (a) control vs. Kabuki syndrome (KS) (AUC = 0.80), (b) control vs. Wiedemann-Steiner syndrome (WSS) (AUC = 0.73), and (c) KS vs. WSS (AUC = 0.85).
  • Figure 5: Attention heatmaps generated by the model for the three classification tasks: (a) control vs. Kabuki syndrome (KS), (b) control vs. Wiedemann-Steiner syndrome (WSS), and (c) KS vs. WSS. Each column displays three representative fingerprint images overlaid with attention heatmaps from participants with the corresponding conditions. Column (a) shows fingerprints from one control participant and two participants with KS; column (b) shows fingerprints from two control participants and one participant with WSS; and column (c) shows fingerprints from one participant with KS and two participants with WSS. Heatmaps were derived from class-token self-attention weights, averaged across all transformer layers and heads, and overlaid on the original fingerprint images, with warmer colors indicating higher attention scores.